xref: /openbmc/linux/fs/nfs/blocklayout/blocklayout.c (revision 80483c3a)
1 /*
2  *  linux/fs/nfs/blocklayout/blocklayout.c
3  *
4  *  Module for the NFSv4.1 pNFS block layout driver.
5  *
6  *  Copyright (c) 2006 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Andy Adamson <andros@citi.umich.edu>
10  *  Fred Isaman <iisaman@umich.edu>
11  *
12  * permission is granted to use, copy, create derivative works and
13  * redistribute this software and such derivative works for any purpose,
14  * so long as the name of the university of michigan is not used in
15  * any advertising or publicity pertaining to the use or distribution
16  * of this software without specific, written prior authorization.  if
17  * the above copyright notice or any other identification of the
18  * university of michigan is included in any copy of any portion of
19  * this software, then the disclaimer below must also be included.
20  *
21  * this software is provided as is, without representation from the
22  * university of michigan as to its fitness for any purpose, and without
23  * warranty by the university of michigan of any kind, either express
24  * or implied, including without limitation the implied warranties of
25  * merchantability and fitness for a particular purpose.  the regents
26  * of the university of michigan shall not be liable for any damages,
27  * including special, indirect, incidental, or consequential damages,
28  * with respect to any claim arising out or in connection with the use
29  * of the software, even if it has been or is hereafter advised of the
30  * possibility of such damages.
31  */
32 
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/bio.h>		/* struct bio */
38 #include <linux/prefetch.h>
39 #include <linux/pagevec.h>
40 
41 #include "../pnfs.h"
42 #include "../nfs4session.h"
43 #include "../internal.h"
44 #include "blocklayout.h"
45 
46 #define NFSDBG_FACILITY	NFSDBG_PNFS_LD
47 
48 MODULE_LICENSE("GPL");
49 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
50 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
51 
52 static bool is_hole(struct pnfs_block_extent *be)
53 {
54 	switch (be->be_state) {
55 	case PNFS_BLOCK_NONE_DATA:
56 		return true;
57 	case PNFS_BLOCK_INVALID_DATA:
58 		return be->be_tag ? false : true;
59 	default:
60 		return false;
61 	}
62 }
63 
64 /* The data we are handed might be spread across several bios.  We need
65  * to track when the last one is finished.
66  */
67 struct parallel_io {
68 	struct kref refcnt;
69 	void (*pnfs_callback) (void *data);
70 	void *data;
71 };
72 
73 static inline struct parallel_io *alloc_parallel(void *data)
74 {
75 	struct parallel_io *rv;
76 
77 	rv  = kmalloc(sizeof(*rv), GFP_NOFS);
78 	if (rv) {
79 		rv->data = data;
80 		kref_init(&rv->refcnt);
81 	}
82 	return rv;
83 }
84 
85 static inline void get_parallel(struct parallel_io *p)
86 {
87 	kref_get(&p->refcnt);
88 }
89 
90 static void destroy_parallel(struct kref *kref)
91 {
92 	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
93 
94 	dprintk("%s enter\n", __func__);
95 	p->pnfs_callback(p->data);
96 	kfree(p);
97 }
98 
99 static inline void put_parallel(struct parallel_io *p)
100 {
101 	kref_put(&p->refcnt, destroy_parallel);
102 }
103 
104 static struct bio *
105 bl_submit_bio(struct bio *bio)
106 {
107 	if (bio) {
108 		get_parallel(bio->bi_private);
109 		dprintk("%s submitting %s bio %u@%llu\n", __func__,
110 			bio_op(bio) == READ ? "read" : "write",
111 			bio->bi_iter.bi_size,
112 			(unsigned long long)bio->bi_iter.bi_sector);
113 		submit_bio(bio);
114 	}
115 	return NULL;
116 }
117 
118 static struct bio *
119 bl_alloc_init_bio(int npg, struct block_device *bdev, sector_t disk_sector,
120 		bio_end_io_t end_io, struct parallel_io *par)
121 {
122 	struct bio *bio;
123 
124 	npg = min(npg, BIO_MAX_PAGES);
125 	bio = bio_alloc(GFP_NOIO, npg);
126 	if (!bio && (current->flags & PF_MEMALLOC)) {
127 		while (!bio && (npg /= 2))
128 			bio = bio_alloc(GFP_NOIO, npg);
129 	}
130 
131 	if (bio) {
132 		bio->bi_iter.bi_sector = disk_sector;
133 		bio->bi_bdev = bdev;
134 		bio->bi_end_io = end_io;
135 		bio->bi_private = par;
136 	}
137 	return bio;
138 }
139 
140 static struct bio *
141 do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect,
142 		struct page *page, struct pnfs_block_dev_map *map,
143 		struct pnfs_block_extent *be, bio_end_io_t end_io,
144 		struct parallel_io *par, unsigned int offset, int *len)
145 {
146 	struct pnfs_block_dev *dev =
147 		container_of(be->be_device, struct pnfs_block_dev, node);
148 	u64 disk_addr, end;
149 
150 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
151 		npg, rw, (unsigned long long)isect, offset, *len);
152 
153 	/* translate to device offset */
154 	isect += be->be_v_offset;
155 	isect -= be->be_f_offset;
156 
157 	/* translate to physical disk offset */
158 	disk_addr = (u64)isect << SECTOR_SHIFT;
159 	if (disk_addr < map->start || disk_addr >= map->start + map->len) {
160 		if (!dev->map(dev, disk_addr, map))
161 			return ERR_PTR(-EIO);
162 		bio = bl_submit_bio(bio);
163 	}
164 	disk_addr += map->disk_offset;
165 	disk_addr -= map->start;
166 
167 	/* limit length to what the device mapping allows */
168 	end = disk_addr + *len;
169 	if (end >= map->start + map->len)
170 		*len = map->start + map->len - disk_addr;
171 
172 retry:
173 	if (!bio) {
174 		bio = bl_alloc_init_bio(npg, map->bdev,
175 				disk_addr >> SECTOR_SHIFT, end_io, par);
176 		if (!bio)
177 			return ERR_PTR(-ENOMEM);
178 		bio_set_op_attrs(bio, rw, 0);
179 	}
180 	if (bio_add_page(bio, page, *len, offset) < *len) {
181 		bio = bl_submit_bio(bio);
182 		goto retry;
183 	}
184 	return bio;
185 }
186 
187 static void bl_end_io_read(struct bio *bio)
188 {
189 	struct parallel_io *par = bio->bi_private;
190 
191 	if (bio->bi_error) {
192 		struct nfs_pgio_header *header = par->data;
193 
194 		if (!header->pnfs_error)
195 			header->pnfs_error = -EIO;
196 		pnfs_set_lo_fail(header->lseg);
197 	}
198 
199 	bio_put(bio);
200 	put_parallel(par);
201 }
202 
203 static void bl_read_cleanup(struct work_struct *work)
204 {
205 	struct rpc_task *task;
206 	struct nfs_pgio_header *hdr;
207 	dprintk("%s enter\n", __func__);
208 	task = container_of(work, struct rpc_task, u.tk_work);
209 	hdr = container_of(task, struct nfs_pgio_header, task);
210 	pnfs_ld_read_done(hdr);
211 }
212 
213 static void
214 bl_end_par_io_read(void *data)
215 {
216 	struct nfs_pgio_header *hdr = data;
217 
218 	hdr->task.tk_status = hdr->pnfs_error;
219 	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
220 	schedule_work(&hdr->task.u.tk_work);
221 }
222 
223 static enum pnfs_try_status
224 bl_read_pagelist(struct nfs_pgio_header *header)
225 {
226 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
227 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
228 	struct bio *bio = NULL;
229 	struct pnfs_block_extent be;
230 	sector_t isect, extent_length = 0;
231 	struct parallel_io *par;
232 	loff_t f_offset = header->args.offset;
233 	size_t bytes_left = header->args.count;
234 	unsigned int pg_offset = header->args.pgbase, pg_len;
235 	struct page **pages = header->args.pages;
236 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
237 	const bool is_dio = (header->dreq != NULL);
238 	struct blk_plug plug;
239 	int i;
240 
241 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
242 		header->page_array.npages, f_offset,
243 		(unsigned int)header->args.count);
244 
245 	par = alloc_parallel(header);
246 	if (!par)
247 		return PNFS_NOT_ATTEMPTED;
248 	par->pnfs_callback = bl_end_par_io_read;
249 
250 	blk_start_plug(&plug);
251 
252 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
253 	/* Code assumes extents are page-aligned */
254 	for (i = pg_index; i < header->page_array.npages; i++) {
255 		if (extent_length <= 0) {
256 			/* We've used up the previous extent */
257 			bio = bl_submit_bio(bio);
258 
259 			/* Get the next one */
260 			if (!ext_tree_lookup(bl, isect, &be, false)) {
261 				header->pnfs_error = -EIO;
262 				goto out;
263 			}
264 			extent_length = be.be_length - (isect - be.be_f_offset);
265 		}
266 
267 		if (is_dio) {
268 			if (pg_offset + bytes_left > PAGE_SIZE)
269 				pg_len = PAGE_SIZE - pg_offset;
270 			else
271 				pg_len = bytes_left;
272 		} else {
273 			BUG_ON(pg_offset != 0);
274 			pg_len = PAGE_SIZE;
275 		}
276 
277 		if (is_hole(&be)) {
278 			bio = bl_submit_bio(bio);
279 			/* Fill hole w/ zeroes w/o accessing device */
280 			dprintk("%s Zeroing page for hole\n", __func__);
281 			zero_user_segment(pages[i], pg_offset, pg_len);
282 
283 			/* invalidate map */
284 			map.start = NFS4_MAX_UINT64;
285 		} else {
286 			bio = do_add_page_to_bio(bio,
287 						 header->page_array.npages - i,
288 						 READ,
289 						 isect, pages[i], &map, &be,
290 						 bl_end_io_read, par,
291 						 pg_offset, &pg_len);
292 			if (IS_ERR(bio)) {
293 				header->pnfs_error = PTR_ERR(bio);
294 				bio = NULL;
295 				goto out;
296 			}
297 		}
298 		isect += (pg_len >> SECTOR_SHIFT);
299 		extent_length -= (pg_len >> SECTOR_SHIFT);
300 		f_offset += pg_len;
301 		bytes_left -= pg_len;
302 		pg_offset = 0;
303 	}
304 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
305 		header->res.eof = 1;
306 		header->res.count = header->inode->i_size - header->args.offset;
307 	} else {
308 		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
309 	}
310 out:
311 	bl_submit_bio(bio);
312 	blk_finish_plug(&plug);
313 	put_parallel(par);
314 	return PNFS_ATTEMPTED;
315 }
316 
317 static void bl_end_io_write(struct bio *bio)
318 {
319 	struct parallel_io *par = bio->bi_private;
320 	struct nfs_pgio_header *header = par->data;
321 
322 	if (bio->bi_error) {
323 		if (!header->pnfs_error)
324 			header->pnfs_error = -EIO;
325 		pnfs_set_lo_fail(header->lseg);
326 	}
327 	bio_put(bio);
328 	put_parallel(par);
329 }
330 
331 /* Function scheduled for call during bl_end_par_io_write,
332  * it marks sectors as written and extends the commitlist.
333  */
334 static void bl_write_cleanup(struct work_struct *work)
335 {
336 	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
337 	struct nfs_pgio_header *hdr =
338 			container_of(task, struct nfs_pgio_header, task);
339 
340 	dprintk("%s enter\n", __func__);
341 
342 	if (likely(!hdr->pnfs_error)) {
343 		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
344 		u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
345 		u64 end = (hdr->args.offset + hdr->args.count +
346 			PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
347 
348 		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
349 					(end - start) >> SECTOR_SHIFT);
350 	}
351 
352 	pnfs_ld_write_done(hdr);
353 }
354 
355 /* Called when last of bios associated with a bl_write_pagelist call finishes */
356 static void bl_end_par_io_write(void *data)
357 {
358 	struct nfs_pgio_header *hdr = data;
359 
360 	hdr->task.tk_status = hdr->pnfs_error;
361 	hdr->verf.committed = NFS_FILE_SYNC;
362 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
363 	schedule_work(&hdr->task.u.tk_work);
364 }
365 
366 static enum pnfs_try_status
367 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
368 {
369 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
370 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
371 	struct bio *bio = NULL;
372 	struct pnfs_block_extent be;
373 	sector_t isect, extent_length = 0;
374 	struct parallel_io *par = NULL;
375 	loff_t offset = header->args.offset;
376 	size_t count = header->args.count;
377 	struct page **pages = header->args.pages;
378 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
379 	unsigned int pg_len;
380 	struct blk_plug plug;
381 	int i;
382 
383 	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
384 
385 	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
386 	 * We want to write each, and if there is an error set pnfs_error
387 	 * to have it redone using nfs.
388 	 */
389 	par = alloc_parallel(header);
390 	if (!par)
391 		return PNFS_NOT_ATTEMPTED;
392 	par->pnfs_callback = bl_end_par_io_write;
393 
394 	blk_start_plug(&plug);
395 
396 	/* we always write out the whole page */
397 	offset = offset & (loff_t)PAGE_MASK;
398 	isect = offset >> SECTOR_SHIFT;
399 
400 	for (i = pg_index; i < header->page_array.npages; i++) {
401 		if (extent_length <= 0) {
402 			/* We've used up the previous extent */
403 			bio = bl_submit_bio(bio);
404 			/* Get the next one */
405 			if (!ext_tree_lookup(bl, isect, &be, true)) {
406 				header->pnfs_error = -EINVAL;
407 				goto out;
408 			}
409 
410 			extent_length = be.be_length - (isect - be.be_f_offset);
411 		}
412 
413 		pg_len = PAGE_SIZE;
414 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
415 					 WRITE, isect, pages[i], &map, &be,
416 					 bl_end_io_write, par,
417 					 0, &pg_len);
418 		if (IS_ERR(bio)) {
419 			header->pnfs_error = PTR_ERR(bio);
420 			bio = NULL;
421 			goto out;
422 		}
423 
424 		offset += pg_len;
425 		count -= pg_len;
426 		isect += (pg_len >> SECTOR_SHIFT);
427 		extent_length -= (pg_len >> SECTOR_SHIFT);
428 	}
429 
430 	header->res.count = header->args.count;
431 out:
432 	bl_submit_bio(bio);
433 	blk_finish_plug(&plug);
434 	put_parallel(par);
435 	return PNFS_ATTEMPTED;
436 }
437 
438 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
439 {
440 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
441 	int err;
442 
443 	dprintk("%s enter\n", __func__);
444 
445 	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
446 	WARN_ON(err);
447 
448 	kfree(bl);
449 }
450 
451 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
452 		gfp_t gfp_flags, bool is_scsi_layout)
453 {
454 	struct pnfs_block_layout *bl;
455 
456 	dprintk("%s enter\n", __func__);
457 	bl = kzalloc(sizeof(*bl), gfp_flags);
458 	if (!bl)
459 		return NULL;
460 
461 	bl->bl_ext_rw = RB_ROOT;
462 	bl->bl_ext_ro = RB_ROOT;
463 	spin_lock_init(&bl->bl_ext_lock);
464 
465 	bl->bl_scsi_layout = is_scsi_layout;
466 	return &bl->bl_layout;
467 }
468 
469 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
470 						   gfp_t gfp_flags)
471 {
472 	return __bl_alloc_layout_hdr(inode, gfp_flags, false);
473 }
474 
475 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
476 						   gfp_t gfp_flags)
477 {
478 	return __bl_alloc_layout_hdr(inode, gfp_flags, true);
479 }
480 
481 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
482 {
483 	dprintk("%s enter\n", __func__);
484 	kfree(lseg);
485 }
486 
487 /* Tracks info needed to ensure extents in layout obey constraints of spec */
488 struct layout_verification {
489 	u32 mode;	/* R or RW */
490 	u64 start;	/* Expected start of next non-COW extent */
491 	u64 inval;	/* Start of INVAL coverage */
492 	u64 cowread;	/* End of COW read coverage */
493 };
494 
495 /* Verify the extent meets the layout requirements of the pnfs-block draft,
496  * section 2.3.1.
497  */
498 static int verify_extent(struct pnfs_block_extent *be,
499 			 struct layout_verification *lv)
500 {
501 	if (lv->mode == IOMODE_READ) {
502 		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
503 		    be->be_state == PNFS_BLOCK_INVALID_DATA)
504 			return -EIO;
505 		if (be->be_f_offset != lv->start)
506 			return -EIO;
507 		lv->start += be->be_length;
508 		return 0;
509 	}
510 	/* lv->mode == IOMODE_RW */
511 	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
512 		if (be->be_f_offset != lv->start)
513 			return -EIO;
514 		if (lv->cowread > lv->start)
515 			return -EIO;
516 		lv->start += be->be_length;
517 		lv->inval = lv->start;
518 		return 0;
519 	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
520 		if (be->be_f_offset != lv->start)
521 			return -EIO;
522 		lv->start += be->be_length;
523 		return 0;
524 	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
525 		if (be->be_f_offset > lv->start)
526 			return -EIO;
527 		if (be->be_f_offset < lv->inval)
528 			return -EIO;
529 		if (be->be_f_offset < lv->cowread)
530 			return -EIO;
531 		/* It looks like you might want to min this with lv->start,
532 		 * but you really don't.
533 		 */
534 		lv->inval = lv->inval + be->be_length;
535 		lv->cowread = be->be_f_offset + be->be_length;
536 		return 0;
537 	} else
538 		return -EIO;
539 }
540 
541 static int decode_sector_number(__be32 **rp, sector_t *sp)
542 {
543 	uint64_t s;
544 
545 	*rp = xdr_decode_hyper(*rp, &s);
546 	if (s & 0x1ff) {
547 		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
548 		return -1;
549 	}
550 	*sp = s >> SECTOR_SHIFT;
551 	return 0;
552 }
553 
554 static int
555 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
556 		struct layout_verification *lv, struct list_head *extents,
557 		gfp_t gfp_mask)
558 {
559 	struct pnfs_block_extent *be;
560 	struct nfs4_deviceid id;
561 	int error;
562 	__be32 *p;
563 
564 	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
565 	if (!p)
566 		return -EIO;
567 
568 	be = kzalloc(sizeof(*be), GFP_NOFS);
569 	if (!be)
570 		return -ENOMEM;
571 
572 	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
573 	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
574 
575 	error = -EIO;
576 	be->be_device = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
577 						lo->plh_lc_cred, gfp_mask);
578 	if (!be->be_device)
579 		goto out_free_be;
580 
581 	/*
582 	 * The next three values are read in as bytes, but stored in the
583 	 * extent structure in 512-byte granularity.
584 	 */
585 	if (decode_sector_number(&p, &be->be_f_offset) < 0)
586 		goto out_put_deviceid;
587 	if (decode_sector_number(&p, &be->be_length) < 0)
588 		goto out_put_deviceid;
589 	if (decode_sector_number(&p, &be->be_v_offset) < 0)
590 		goto out_put_deviceid;
591 	be->be_state = be32_to_cpup(p++);
592 
593 	error = verify_extent(be, lv);
594 	if (error) {
595 		dprintk("%s: extent verification failed\n", __func__);
596 		goto out_put_deviceid;
597 	}
598 
599 	list_add_tail(&be->be_list, extents);
600 	return 0;
601 
602 out_put_deviceid:
603 	nfs4_put_deviceid_node(be->be_device);
604 out_free_be:
605 	kfree(be);
606 	return error;
607 }
608 
609 static struct pnfs_layout_segment *
610 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
611 		gfp_t gfp_mask)
612 {
613 	struct layout_verification lv = {
614 		.mode = lgr->range.iomode,
615 		.start = lgr->range.offset >> SECTOR_SHIFT,
616 		.inval = lgr->range.offset >> SECTOR_SHIFT,
617 		.cowread = lgr->range.offset >> SECTOR_SHIFT,
618 	};
619 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
620 	struct pnfs_layout_segment *lseg;
621 	struct xdr_buf buf;
622 	struct xdr_stream xdr;
623 	struct page *scratch;
624 	int status, i;
625 	uint32_t count;
626 	__be32 *p;
627 	LIST_HEAD(extents);
628 
629 	dprintk("---> %s\n", __func__);
630 
631 	lseg = kzalloc(sizeof(*lseg), gfp_mask);
632 	if (!lseg)
633 		return ERR_PTR(-ENOMEM);
634 
635 	status = -ENOMEM;
636 	scratch = alloc_page(gfp_mask);
637 	if (!scratch)
638 		goto out;
639 
640 	xdr_init_decode_pages(&xdr, &buf,
641 			lgr->layoutp->pages, lgr->layoutp->len);
642 	xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
643 
644 	status = -EIO;
645 	p = xdr_inline_decode(&xdr, 4);
646 	if (unlikely(!p))
647 		goto out_free_scratch;
648 
649 	count = be32_to_cpup(p++);
650 	dprintk("%s: number of extents %d\n", __func__, count);
651 
652 	/*
653 	 * Decode individual extents, putting them in temporary staging area
654 	 * until whole layout is decoded to make error recovery easier.
655 	 */
656 	for (i = 0; i < count; i++) {
657 		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
658 		if (status)
659 			goto process_extents;
660 	}
661 
662 	if (lgr->range.offset + lgr->range.length !=
663 			lv.start << SECTOR_SHIFT) {
664 		dprintk("%s Final length mismatch\n", __func__);
665 		status = -EIO;
666 		goto process_extents;
667 	}
668 
669 	if (lv.start < lv.cowread) {
670 		dprintk("%s Final uncovered COW extent\n", __func__);
671 		status = -EIO;
672 	}
673 
674 process_extents:
675 	while (!list_empty(&extents)) {
676 		struct pnfs_block_extent *be =
677 			list_first_entry(&extents, struct pnfs_block_extent,
678 					 be_list);
679 		list_del(&be->be_list);
680 
681 		if (!status)
682 			status = ext_tree_insert(bl, be);
683 
684 		if (status) {
685 			nfs4_put_deviceid_node(be->be_device);
686 			kfree(be);
687 		}
688 	}
689 
690 out_free_scratch:
691 	__free_page(scratch);
692 out:
693 	dprintk("%s returns %d\n", __func__, status);
694 	if (status) {
695 		kfree(lseg);
696 		return ERR_PTR(status);
697 	}
698 	return lseg;
699 }
700 
701 static void
702 bl_return_range(struct pnfs_layout_hdr *lo,
703 		struct pnfs_layout_range *range)
704 {
705 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
706 	sector_t offset = range->offset >> SECTOR_SHIFT, end;
707 
708 	if (range->offset % 8) {
709 		dprintk("%s: offset %lld not block size aligned\n",
710 			__func__, range->offset);
711 		return;
712 	}
713 
714 	if (range->length != NFS4_MAX_UINT64) {
715 		if (range->length % 8) {
716 			dprintk("%s: length %lld not block size aligned\n",
717 				__func__, range->length);
718 			return;
719 		}
720 
721 		end = offset + (range->length >> SECTOR_SHIFT);
722 	} else {
723 		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
724 	}
725 
726 	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
727 }
728 
729 static int
730 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
731 {
732 	return ext_tree_prepare_commit(arg);
733 }
734 
735 static void
736 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
737 {
738 	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
739 }
740 
741 static int
742 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
743 {
744 	dprintk("%s enter\n", __func__);
745 
746 	if (server->pnfs_blksize == 0) {
747 		dprintk("%s Server did not return blksize\n", __func__);
748 		return -EINVAL;
749 	}
750 	if (server->pnfs_blksize > PAGE_SIZE) {
751 		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
752 			__func__, server->pnfs_blksize);
753 		return -EINVAL;
754 	}
755 
756 	return 0;
757 }
758 
759 static bool
760 is_aligned_req(struct nfs_pageio_descriptor *pgio,
761 		struct nfs_page *req, unsigned int alignment, bool is_write)
762 {
763 	/*
764 	 * Always accept buffered writes, higher layers take care of the
765 	 * right alignment.
766 	 */
767 	if (pgio->pg_dreq == NULL)
768 		return true;
769 
770 	if (!IS_ALIGNED(req->wb_offset, alignment))
771 		return false;
772 
773 	if (IS_ALIGNED(req->wb_bytes, alignment))
774 		return true;
775 
776 	if (is_write &&
777 	    (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
778 		/*
779 		 * If the write goes up to the inode size, just write
780 		 * the full page.  Data past the inode size is
781 		 * guaranteed to be zeroed by the higher level client
782 		 * code, and this behaviour is mandated by RFC 5663
783 		 * section 2.3.2.
784 		 */
785 		return true;
786 	}
787 
788 	return false;
789 }
790 
791 static void
792 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
793 {
794 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
795 		nfs_pageio_reset_read_mds(pgio);
796 		return;
797 	}
798 
799 	pnfs_generic_pg_init_read(pgio, req);
800 }
801 
802 /*
803  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
804  * of bytes (maximum @req->wb_bytes) that can be coalesced.
805  */
806 static size_t
807 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
808 		struct nfs_page *req)
809 {
810 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
811 		return 0;
812 	return pnfs_generic_pg_test(pgio, prev, req);
813 }
814 
815 /*
816  * Return the number of contiguous bytes for a given inode
817  * starting at page frame idx.
818  */
819 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
820 {
821 	struct address_space *mapping = inode->i_mapping;
822 	pgoff_t end;
823 
824 	/* Optimize common case that writes from 0 to end of file */
825 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
826 	if (end != inode->i_mapping->nrpages) {
827 		rcu_read_lock();
828 		end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
829 		rcu_read_unlock();
830 	}
831 
832 	if (!end)
833 		return i_size_read(inode) - (idx << PAGE_SHIFT);
834 	else
835 		return (end - idx) << PAGE_SHIFT;
836 }
837 
838 static void
839 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
840 {
841 	u64 wb_size;
842 
843 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
844 		nfs_pageio_reset_write_mds(pgio);
845 		return;
846 	}
847 
848 	if (pgio->pg_dreq == NULL)
849 		wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
850 					      req->wb_index);
851 	else
852 		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
853 
854 	pnfs_generic_pg_init_write(pgio, req, wb_size);
855 }
856 
857 /*
858  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
859  * of bytes (maximum @req->wb_bytes) that can be coalesced.
860  */
861 static size_t
862 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
863 		 struct nfs_page *req)
864 {
865 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
866 		return 0;
867 	return pnfs_generic_pg_test(pgio, prev, req);
868 }
869 
870 static const struct nfs_pageio_ops bl_pg_read_ops = {
871 	.pg_init = bl_pg_init_read,
872 	.pg_test = bl_pg_test_read,
873 	.pg_doio = pnfs_generic_pg_readpages,
874 	.pg_cleanup = pnfs_generic_pg_cleanup,
875 };
876 
877 static const struct nfs_pageio_ops bl_pg_write_ops = {
878 	.pg_init = bl_pg_init_write,
879 	.pg_test = bl_pg_test_write,
880 	.pg_doio = pnfs_generic_pg_writepages,
881 	.pg_cleanup = pnfs_generic_pg_cleanup,
882 };
883 
884 static struct pnfs_layoutdriver_type blocklayout_type = {
885 	.id				= LAYOUT_BLOCK_VOLUME,
886 	.name				= "LAYOUT_BLOCK_VOLUME",
887 	.owner				= THIS_MODULE,
888 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
889 					  PNFS_READ_WHOLE_PAGE,
890 	.read_pagelist			= bl_read_pagelist,
891 	.write_pagelist			= bl_write_pagelist,
892 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
893 	.free_layout_hdr		= bl_free_layout_hdr,
894 	.alloc_lseg			= bl_alloc_lseg,
895 	.free_lseg			= bl_free_lseg,
896 	.return_range			= bl_return_range,
897 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
898 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
899 	.set_layoutdriver		= bl_set_layoutdriver,
900 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
901 	.free_deviceid_node		= bl_free_deviceid_node,
902 	.pg_read_ops			= &bl_pg_read_ops,
903 	.pg_write_ops			= &bl_pg_write_ops,
904 	.sync				= pnfs_generic_sync,
905 };
906 
907 static struct pnfs_layoutdriver_type scsilayout_type = {
908 	.id				= LAYOUT_SCSI,
909 	.name				= "LAYOUT_SCSI",
910 	.owner				= THIS_MODULE,
911 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
912 					  PNFS_READ_WHOLE_PAGE,
913 	.read_pagelist			= bl_read_pagelist,
914 	.write_pagelist			= bl_write_pagelist,
915 	.alloc_layout_hdr		= sl_alloc_layout_hdr,
916 	.free_layout_hdr		= bl_free_layout_hdr,
917 	.alloc_lseg			= bl_alloc_lseg,
918 	.free_lseg			= bl_free_lseg,
919 	.return_range			= bl_return_range,
920 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
921 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
922 	.set_layoutdriver		= bl_set_layoutdriver,
923 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
924 	.free_deviceid_node		= bl_free_deviceid_node,
925 	.pg_read_ops			= &bl_pg_read_ops,
926 	.pg_write_ops			= &bl_pg_write_ops,
927 	.sync				= pnfs_generic_sync,
928 };
929 
930 
931 static int __init nfs4blocklayout_init(void)
932 {
933 	int ret;
934 
935 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
936 
937 	ret = bl_init_pipefs();
938 	if (ret)
939 		goto out;
940 
941 	ret = pnfs_register_layoutdriver(&blocklayout_type);
942 	if (ret)
943 		goto out_cleanup_pipe;
944 
945 	ret = pnfs_register_layoutdriver(&scsilayout_type);
946 	if (ret)
947 		goto out_unregister_block;
948 	return 0;
949 
950 out_unregister_block:
951 	pnfs_unregister_layoutdriver(&blocklayout_type);
952 out_cleanup_pipe:
953 	bl_cleanup_pipefs();
954 out:
955 	return ret;
956 }
957 
958 static void __exit nfs4blocklayout_exit(void)
959 {
960 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
961 	       __func__);
962 
963 	pnfs_unregister_layoutdriver(&scsilayout_type);
964 	pnfs_unregister_layoutdriver(&blocklayout_type);
965 	bl_cleanup_pipefs();
966 }
967 
968 MODULE_ALIAS("nfs-layouttype4-3");
969 
970 module_init(nfs4blocklayout_init);
971 module_exit(nfs4blocklayout_exit);
972